---
license: apache-2.0
tags:
- time series
- forecasting
- pretrained models
- foundation models
- time series foundation models
---

# Lag-Llama: Towards Foundation Models for Probabilistic Time Series Forecasting

![lag-llama-architecture](images/lagllama.webp)

Lag-Llama is the <b>first open-source foundation model for time series forecasting</b>!

Twitter Thread: https://twitter.com.

HuggingFace: {}

Colab Demo: {}

Paper: {Not arxiv}.

arXiv has a previous outdated version of the paper and is still being updated with the latest version; please use the above link to access the latest version.

This repository houses the Lag-Llama architecture.

<b>Current Features:</b>

1. <b>Zero-shot forecasting</b> on a dataset of <b>any frequency</b> for <b>any prediction length</b>, using the Colab Demo.

Coming Soon:

1. An <b>online gradio demo</b> to upload time series and get zero-shot predictions for
1. Features for <b>finetuning</b> the foundation model
2. Features for <b>pretraining</b> Lag-Llama on your own large-scale data
3. Scripts to <b>reproduce</b> all results in the paper.